Ep. 11: WTF is a Custom GPT?

[00:00:00] Foreign. [00:00:05] Welcome to ChatGPT Curious, a podcast for people who are, well, curious about ChatGPT. I'm, um, your host, Dr. Shantae Cofield, also known as the Maestro, and I created this show to explore what ChatGPT actually is. Really, though, are the files in the computer, how to use it, and what it might mean for how we think, work, create, and move through life. Whether you're skeptical, intrigued, or already experimenting, you're in the right place. All that I ask is that you stay curious. All right, let's get into it. [00:00:38] Hello, hello, hello, my curious people, and welcome to episode 11 of ChatGPT Curious. I am your most grateful host, the Maestro, and today we are talking about custom GPTs. If you've listened to every episode thus far. Thank you. Uh, but you may have heard me circle the drain with this topic. I kind of talking about it, alluding to it, I was like, I don't really understand the utility of this. Uh, but today we are jumping all the way in because I finally jumped all the way in and I built one myself. So we are answering the question, what the is a custom GPT? So real quick, before we hop on in, two things. Number one, welcome to October. Uh, I do like to add a bit of timeliness to these somewhat evergreen episodes. So at Evergreen, meaning they're kind of like always relevant. [00:01:23] Uh, that is, until the tech changes. But, you know, you can listen to this thing anytime and it's still gonna be relevant. Um, but this episode is gonna drop on October 2nd, so if you're listening to it on the day that drops, welcome to October. [00:01:36] And, uh, number two, thank you for the dope convos and reviews. Like, I had a really, really, really, really good convo with why I'm saying com, though, really great conversation with the homie Kate. [00:01:51] Um, she also left a dope ass review. Thank you. And I'm actually going to circle back to her at the end of this episode and share how she's been using chat gbt. Um, but, you know, podcasting is a very unidirectional process. If you listen to my Master on the Mic episodes that podcast, rather I say this all the time. Um, it's a very unidirectional process. You're just like talking at the screen and I do love that. Um, but it is really, really dope to hear from listeners and to hear that things are helpful and that you're curious. [00:02:19] Uh, so, so thank you for sharing that with me. And yes, that is my way of encouraging more of. So, uh, custom GPTs. Let's get into it. All right, As a refresher, right? GPT, chat GPT and we're talking about custom GPTs. GPT stands for generative, pre trained transformer. You can head back to episode one for the full refresher. I know that episode was heavy. It had a lot people's head fell off. Go back and listen to it because I bet you will understand more now. [00:02:47] Um, but go ahead, go check out episode episode one. But generative, let's, let's define and talk about each of these words. Generative. That means that it doesn't have to. It doesn't just look things up. It's generating new text, pre trained before it is released to the public. It is trained on massive amounts of likely illegally obtained text where it quote unquote learns patterns, right, of how words connect. And then the new word for us that we've yet to go over, transformer, generative, pre trained, transformer, Transformer speaks to what the model is actually doing in order to generate that output, right? Here we go. Gets a little bit deep. Now it's transforming vectors. [00:03:33] So take a breath folks. [00:03:37] Okay, here we go. Natural language, right? With the stuff that we type in to the, to the prompt box, tom field, whatever you want to call it. Natural language that gets broken down into tokens, right? Tokens are chunks of characters. Again, we talked about this in episode one. You can go back to that. [00:03:52] We type something in, it gets broken down internally into these chunks of characters that are called tokens. Tokens are represented by a series of numbers. And this series of numbers is called a vector, the model. It then takes each vector and it transforms it into another vector, right? It goes from one set of numbers, one series of numbers, to another series of numbers. And this next second series of numbers better represents and expresses how that vector is related to the other words around it, right? How that chunk of text is related to the other text that's around it, right? This is how the word bank can mean money in one sentence and riverside in another sentence. This is how it understands that quote unquote understands, right? The vector, the series of numbers is being transformed based on context. [00:04:54] This transforming it happens many times based on the number of what's called transformer layers that a model has. [00:05:01] And then from here we will get an output. We talked about this before. We get an output that is based on probability and the model predicting the most likely next token. I know that it's a little confusing, but again, summary, there it is, all math and it's really cool and Complex math. And if you're curious, feel free to hit rewind and listen again. [00:05:25] Uh, maybe one day I'll do a deeper dive on this. I get curious about it. And so this was actually pretty fun for me to sit with and sit with ChatGPT, as is how I learned this stuff and understand what vector meant and what transform actually meant. And I hear it in some of the YouTube videos I watch those MIT classes and things and I'm like, what the fuck? And so to sit and like kind of sit with it, um, and go through it, I'm like, okay, this, this, this is cool, right? So we're going to move on because I know that you're like, you've already lost me. We're five minutes in my show. You lost me. So we're gonna move on. Uh, but GPT means generative pre trained transformer. Generative meaning generating new text pre trained, meaning it's trained before it reaches the public. And then we just use it, right? It's all locked in and then we just get. [00:06:05] And then transformer, meaning the model transforms the inputs to the most accurately. [00:06:10] Um, excuse me. The model transforms the inputs to most accurately represent relationships between words so that it can generate the output that is most likely to be correct. Right. I try to be pretty specific with the language I use because it does matter. And I do think that over time, as we move towards more advanced, you know, AI using words like understand and reasoning, that is important. That's a lot of marketing speak there. And it's like, does it actually understand? Does it actually know? Is it actually reasoning? Probably not, right? So I tried to be specific and intentional and accurate, um, with these words. So that's why some of this gets a little bit wordy. But I do think in the long run it is, it is important to, to lean on that. So now that your brain is broken, let's talk about a custom GPT. So a custom GPT is basically your own version of version, uh, that sounds like virgin, but I didn't say that version of Chat GPT that works based on the instructions that you give it. If folks, you have ever said, I want to clone myself, this is how you do it. All right? And namely when you're saying I want to clone myself so that like, people keep asking me these questions, I wish I could just call myself, this is how you do it. This right here is how you do it. Get excited. So I titled this episode what the is Custom. What the is a custom GPT Intentionally. I did not title it how to Build a custom GPT. To that end, I'm going to kind of COVID how to build one. But for the deep dive on how to create one, how to build one, I'm going to refer you to the Help Center. I talked about that resource last week. Maybe last week's episode, um, then the week's escape. Maybe it was two weeks ago. Two weeks ago, I think the episode that was like, features you may not know about. Um, so I will link that in the show notes. But if you go to the, the FAQ, um, in the OpenAI help center about GPTs, I know there's a lot of acronyms there, but it's in there and I will link it for you. Just go to the show notes. Okay, so what we're really going to focus on is what it is, what is a custom GPT. So I'm going to start off with how I use it, what I did, how, like what this thing is that I made so you can get a better understanding of what the fucking custom GPT is. Because I feel like it can kind of be nebulous, like, kind of nebulous topic. Like, wait, what, what's going on? So I'm just going to tell you how I used it, what I did, and I think it'll give you a good representation of like. Oh, I understand. [00:08:33] So last week I quote, unquote, built a custom GPT for my Instagram Intensive, right? I'm using built in air quotes, you can't see me, but I put you. Hopefully I emphasized it as having air quotes because it doesn't require any special coding or anything like that. Which, like I've said in previous episodes, is the real magic of LLMs, right? ChatGPT is a type of LLM large language model. Like, the real magic is that they allow you to execute tasks by just inputting natural language, just typing. That's dope, right? So last week I quote, unquote, built this custom GPT for my Instagram Intensive. My Instagram Intensive is a six week online group coaching program that I run and it teaches health and fitness pros exactly how to use Instagram for online business. I am calling this custom GPT that I made my chat GPT A. All right? Chat GPT A. You might understand my goal for this thing. Slashes utility in understanding it that way. They're hearing the name, right? It's the ta. All right? That's what I. That is how I'm using it for this course. The goal of this custom GPT, the reason I made it, is that if my students, people enrolled in the program have a question about literally anything related to the program, all right, any question that they would ask me, they can just ask the. The GPT. The custom GPT. [00:09:49] Yes, I'm going to be there, but for the times that I'm not with them, we have a Facebook group like that. But I'm like, yo, you have this, you have access to this thing. You don't have to wait until next week, you don't have to put it in the Facebook group and wait till I see it, you can ask it right away, right? So if it wants, if they want to ask logistics, which I think is like less of an issue, but if they want to ask IT logistics about the course, they can. If they want to ask for feedback on content, if they want help or ideas for creating content, if they want to ask the custom GPT anything, right, it will give them the same answer that I would give them, right? In my voice, in my tone, with my values. And that's fucking dope. [00:10:24] But how? [00:10:26] Because I gave it the instructions, I uploaded the files that will serve as its knowledge base and I set the capabilities that it has. Again, no coding needed. That's amazing. OpenAI actually makes it really easy to build one. [00:10:42] Uh, they have what's called the GPT Builder. There are two modes in the builder, there's Create and Configure. And in the Create, like, it's like a toggle at the top. In the Create mode, you can just have a conversation with Chat GBD and it'll walk you through what you need and like how to build it, which is amazing. The Configure mode, like the, when you toggle it over, that's what you actually input the information yourself. Um, and I will link that as well. It'll take you directly to the builder. Pretty amazing. Um, I use the configure side of things, but I'm going to walk you through kind of how I did it and just the overview of this, right? So how good this custom GPT is, right? AKA how much it sounds like you, how correct the outputs that it generates, how correct they are. That is fully based on how good of instructions you give it and the quality of the files that you upload. Right? We've been saying this every single episode. The better the input, the better the output. So for my intensive, I teach from outlines, right? Uh, they are very, very in depth outlines. I don't stray from them. Everything is on them. [00:11:45] Everything that I'm going to say and teach in that program is in those outlines. So to create my custom GPT, I uploaded every single call outline, right? I also uploaded a file that has all of the copy from the sales pages. So like all the logistics and things like that are on or in that file as well. [00:12:04] I also uploaded, uh, some of my sales emails because the sales emails capture how I teach, how I write, how I speak, what I value and what I emphasize about the intensive. [00:12:15] Circling back to what I said this a little bit ago, I didn't use uh, the builder to create this custom GPT. I actually made a project first inside of ChatGPT and I uploaded all of those documents that I just said, I uploaded all of those to the project. So basically I looked, I looked to create what was called like, we'd call like a sandbox and in the, in the kind of developer world, but I looked to create an environment that knew about me and knew about my intensive. And then I wanted to have that thing help me create this custom GPT, right? So that's why I did it this way. [00:12:47] Um, I will link the episode that's episode six where I talk about projects, right? [00:12:54] So inside of this project that I made first I started a conversation with Chat GPT and I was like, hey, I've uploaded all these files and I made it memory. I made it project memory only for the, for the project, right? So it can only access things that I've uploaded. It's not like going to the other parts of, of ChatGPT. [00:13:13] So I started the conversation with ChatGPT and I said, uh, within this project. And I said, you know, I want to build this custom GPT for this program and I asked it to help me out. I said, can you give me some instructions, let me know what to think about, uh, let me know what, you know, I need in order to create this custom GPT and help me build it. And I imagine this is like what happens inside of the builder as well. But I wasn't aware that the builder existed when uh, I was building it. So I was like, I'm just gonna ask it how to, how to do it. And the lesson here that I, that I try to hammer home as in as many episodes as possible, is always ask Chat GPT to help you and to tell you how to use ChatGPT. It's amazing. [00:13:50] So if we zoom out for a second, that's how my brain works. The goal here, if you want to create the best custom GPT, right where it gives the best, most correct answers, it sounds like you, it's got your vibe the, the way that you do that is to upload files that it can read so that it knows the content, right? Stuff that you're talking about. [00:14:10] And if you want it to be a clone of you, want it to sound like you, you need to upload documents so that it knows you, right? So to know the content that you want it to be helping you with and you know, the, the students and the people, whoever are going to be asking about you, just upload files that contain that content. It's very simple. [00:14:29] So obviously if your program is older, if you have a lot of reps with things, this is much easier because you already generated all this stuff. Like I, I am seventh. This will be my 17th round of the intensive. I know this thing in and out. I know the values. I know my values. I know what the, the purpose of this thing is. I know this thing just m. Like, this is my right. If you're just starting with a program, it could be a little harder because you're like, I think that maybe they might have these questions. I think this is like, I want to emphasize this, but I know and I have all the content already of the, all the sales page. I have everything. So, right, so I uploaded uh, all of that. So again, for this thing to have mastered the content, right? And for it to know all of the content, you have to upload files that contain the content. Very simple. [00:15:12] For this custom GPT to know you, to sound like you, to be able to be like, yo, I am, um, the best TA in the world. You have to upload files that contain you. [00:15:26] The easiest way in my opinion, to do this, right, to upload, to create these files that contain you. And your essence is to create a file by having Chat GPT interview you. Easy, right? A little hack here is that. So remember, I made a project, clearly I'm excited about this, right? I made a project for this intensive and it had all of the stuff, everything that I've ever done with the intensive, all the, all of the things like all of the, the weekly calls, um, any supporting documents, my emails, the copy from the sales, all these things. So it has like, it has an under baseline understanding of me. [00:16:04] So I had it interview me, but I also had it answer the questions as well. So here's the, the little bit of a hack is that you can be like, hey, I want to create a, a master document that, you know, really portrays and conveys my essence. Can you interview me? Can you interview me one question at a time? [00:16:24] And then when it asks you the question, Instead of you answering, you can say, based on what you know about me, what would you say? What would, how would you answer this question? And then it's going to give you an answer and then you can just correct the answer as needed. Because some of them you're like, ah, that's spot on. Great, right? Uh, because it's pulling from the stuff that you've uploaded already. [00:16:44] So you, from there you have it, ask all the questions. If you're like, hey, I want you to ask me some more question, I feel like you haven't really, like, done a complete job. Cool, that I'll ask you some more questions. From there, you can have it compile all of those answers to those questions into a single document, right? Call it a manifesto, whatever. And then you upload that document to the custom GPT files. Amazing, right? [00:17:06] The other part of creating the best custom GPT is giving it the best instructions. [00:17:14] All right, so we have two, it's like a two pronged approach here. The best context and then the best instructions. In order to create the best instructions, aka prompts, which we've talked about a million times. Ask ChatGPT to write the instructions for you. [00:17:32] Right? If you always want this custom GPT to say some specific thing, right? That, this, then this, this is how you do that. [00:17:40] All right? If you always want the custom GPT to reference a certain document, this is how you get it to do that. It's with the instructions. But this is how you quote, unquote, program it to do that. The tone that this custom GPT has, the voice, the guardrails, all of this goes into the instructions. And my very strong suggestion is to have Chat GPT write those instructions for you. Okay, A little bit more advanced here. [00:18:08] All right, this next step here, but if you go into the Help center, which I said I strongly advise you to do as you're gonna, if you're gonna build out one of these custom GPTs, if you go to the Help center, it talks about, like, best practices for creating a GPT. And one of the things that it talks about, uh, is creating what's called markdown files. And markdown files are just a type of plain text file. And we wanna create these because it's just faster for ChatGPT to reference them, doesn't have all the other bullshit in the file. [00:18:34] So I did do this, and ChatGPT instructed me to do this. And you can ask it, like, how do I create that? And they'll tell you to like, open the text editor, save it as this. [00:18:42] It um, can. It'll actually create the thing that you're going to paste into the markdown file. So, like, it's doing all the heavy lifting. You have to kind of like execute the little steps. Um, but that is just for those of you that are, that are with me. And you're like, oh, I'm understanding. This is exciting. You want to create those markdown files because it is faster for ChatGPT to reference those and read those things. Okay, so the last part, and again, this episode isn't about how to make a custom GPT, but I know I've thrown some stuff in there and I just, I also want to give you a little, Give you as many heads up as I can. [00:19:16] Um, the last part is setting the custom GPT capabilities. You can allow it or disallow it to search the Internet. You can allow it or not to, um, have people be able to upload files. [00:19:31] Um, these are going to be toggle, toggle on, off fields, right? Fields that you can toggle on and off. And you can ask ChatGPT to help you how to decide how to set them based on your specific use case and be like, should I have it on? Should I have it not have it on? And it'll walk you through that. Right? Um, from here, once you have all that stuff done, you hit publish and you get to set who has access to it. It will generate a link and then you can share that link with your people. Uh, that's awesome. I would suggest that you test it, AKA you ask it things and you see what kind of output it gives you before you share the link with. [00:20:06] You know, if you're like, making it for a course, I would say, I would say test it yourself and also have like some, some people that are close to you test it. Um, I did. I had my sister test it. Um, I had my brother test it. I had Lex tested. I had some, um, folks that are in my ecosystem. I was like, hey, I made this thing, like, play around with it. [00:20:24] Uh, and then I can tweak the files and instructions as needed and just be like, is this thing good? And like, sometimes it's tough for you to like, pretend that you're the user. [00:20:33] Uh, so it can be good to just give to someone that would be a user and be like, hey, what do you, what do you think? Um, but I gave it to my brother to try and I actually went back and added a an about me file because I sent it to him and the first question he asked was, what's my favorite Color. And it responded that it didn't know and to contact the support, you know, the support email. And I was like, hey, maybe I should put that in there. I don't know if people are going to be using it in that way, but, like, it'd be. It would be fun to have it actually give an answer to things. [00:21:00] Um, because I actually had my girl Claudia try it. And one of the first things she wrote was, why is the maestro retiring the program? And I was like, these are things that I wouldn't think to quote, ah, unquote, program it to have an answer for. And the answer that it generated was, uh, really good. And you can have this as one of the guardrails, right in the instructions and be like, do not make anything. If you do not know. If it' not in the files that, that are there, the reference files, then say this specific thing, right? And so I do have that in my instructions, um, as well. But I did go back and, you know, after the testing and, and change some stuff. So to summarize, my friends, a custom GPT is basically your own version of ChatGPT that works based on the instructions that you give it. If we zoom out, the goal, if you're looking to create the best custom GPT, right, where it gives the best, most correct answers and it sounds like you and feels like you the best, the way that you do that is to upload files that it can read so that it knows the content of whatever the it is that you're training it on. [00:22:02] And if you want it to be a clone of you, you have to upload files so that it, quote, unquote, knows you, right? The two prongs that we have of creating the best custom GPT are giving it the best context, slash, content, and giving it the best in the best instructions. [00:22:19] Of note, because if you're sitting here being like, how long does this take? It took me a work day to do it. I sat there on a Saturday and I had volleyball on and tv, uh, and I built it then. Well, with that. And I was like, this is a really cool outcome. So it wasn't something that took me a million years. Again, I already had all the documents or you had things made, aside from the kind of manifesto and the I did the about me actually a few days ago, um, but I did it in less than a day, right? So some use cases for this clearly, like, why would you make a custom GPT is to clone yourself. In previous episodes, I have talked about how I have yet to really see a Good internal facing use case. And when I say internal facing, meaning that, uh, only you as the user have access to it, right? Only you as the person who's programmed it have access to it. [00:23:05] Um, especially given that projects now have project only memory. And I still think that largely the case, like the main use for these is for external facing use cases like when you want to share this thing with other people and have other people access it, which it's dope, right? It is perfect for creating a T.A. of sorts. [00:23:28] Uh, which is why I calling it Chat gpta. Right? So how I use chatgpt this week, right? Every episode, if you're new to the show, I include a section where I briefly discuss how I use Chat GPT that week, that day. And clearly this entire episode was how I use Chat TBD this week. But to circle back to the opening part of the episode, um, where I shouted out the homie Kate, um, I want to talk about how she's been using it, right? Kate is a very curious human. She's PhD, she's big brain, just big, you know, curious human, big curiosity. [00:24:03] Uh, so she has been using Chat GPT for certain things and Claude for certain things. And I'm very intentionally sharing this because last, last, last episode, last week, episode, two weeks episode, um, the homie Rachel asked about Chat GPT versus Gemini and I said use whatever the you want. Like use whatever you like. Um, and so here's a use case where someone is using both. And so Kate is using Claude for her full time job as to help her with her full time job as a professor, right? She's writing a ton of curriculum, so it's not her own direct, it's not going to be her own intellectual property. Um, and she uses it to give her outlines, ask it for help brainstorming, having it create artifacts, which I like, didn't know what that meant until she like went on to elaborate and I was like, okay, that's cool. It's called artifacts. So things like documents, templates, guidelines, discussion prompts, pulling out key id, key key ideas. And then she's using Chat GPT for life stuff. So personal development, communication with her partner, frameworks for, you know, how we relate to each other, attachment styles, things like that, um, using it for conflict resolution and understanding each other's perspectives. [00:25:08] Two different use cases, right? That's great. [00:25:12] Not to say that one is better than the other. They both do things like they both have, in my opinion, they both have caught up. Um, and what tends to happen and I think is very much the case for me is this, like, I'm so deep into using one that I'm like, well, I'm not switching over, like, the thought of having to, like, learn that and train it and feel like I've trained. I'm just like, oh, I, I'm here. But I, I always want to expose you folks to people using things in different ways and give you permission that you don't need to try the things and stay curious and be curious. Right. So Kate did have a question. She had a few questions. I love them. Um, but the one that ties into the episode was, uh, how can I have chat GPT have one mode for one thing versus another mode for something else? And she answered her own question, which would be to have a project with project only memory or as ties into this episode, a custom GPT. [00:26:03] What's worth noting here is that custom GPTs don't save individual conversations. [00:26:09] So if you want to keep whatever, you know, y' all talking about in with that, you know, when you use discussion GPT, you have to copy it somewhere else. Which is also worth noting if you are, um, sharing this with students, something like that. Especially people that have been using chat GPT because they're going to think that, like, oh, the conversations will just be saved in there and I can go back to it. No, they go away. They're done. [00:26:29] So if you want to, you know, have them there, you need to copy them. [00:26:33] If you do want it to remember the past conversations and pull from them, then you'd actually want to make a project that uses project only memory instead of making a custom GPT. Right. So we start to see as we are curious and we start to understand what these things do, we understand better, you know, better how to better use them. And so I think this is an extremely, extremely dope use case. Um, I will report back on how it goes with having it as part of the intensive. But so far, um, sharing it with my folks, they're all just like, yo, this is awesome. And I'm like, I know. I do think you have to be a paid user. Like, don't quote me on that. But based on what the little FAQ thing from OpenAI said, I'm pretty sure you got to be a paid user, um, in order to have this and be able to do this. But, like, for $20 a month, you can create a clone of. You can create a bunch of these, or you can create a clone of yourself and send a link and so people can stop asking you all the questions. [00:27:36] That's amazing. Right again, yes. Put guardrails on things. And we, we do see, if we zoom out, we do see, you know, companies trying to use chatbots for customer service and, like, getting things wrong and, like, there's problems with that. But, like, within your own business, I think there's a ton of utility, especially if you are still having communication with people and you're like, hey, use this first. [00:27:58] You know, go through the things and then contact me if you have deeper questions, more questions, you want to elaborate if it doesn't make sense. [00:28:06] I think it's just a great first step. Uh, and so I'm really, really, really excited to, to use it with, with this cohort and I will let you know how it goes. But that is all that I got for you for today. Hopefully you found this episode helpful. If you did consider leaving a rating or a review. If you want to share it, you can. But this is a bit of a techie episode and folks might cry if this is their introduction, their first introduction to the podcast. So maybe just leave a rating. Maybe just leave a review. [00:28:36] Don't, uh, forget. Also, I have a companion newsletter that drops every Thursday that is basically the episode in text format. So if you prefer to read or you just want that written record, join the newsletter fam. You can head to chatgpt curious.com newsletter or just check out the link in the show notes. [00:28:58] As always, endlessly, endlessly, endless, endlessly appreciative for every single one. [00:29:04] Until we chat again next Thursday. [00:29:07] Take your.